Table 4 ConvNet configuration of DenseNet.
From: Deep learning based automated diagnosis of bone metastases with SPECT thoracic bone images
SPECS D121 | |
121 weight layers | |
input (256 × 256 thoracic image) | |
7 × 7 conv. | |
3 × 3 Max pooling | |
Dense Block (1) | \(\left[ {\begin{array}{*{20}l} {1 \times 1\;{\text{ conv}}.} \\ {3 \times 3 \, \;{\text{conv}}.} \\ \end{array} } \right] \times 6\) |
1 × 1 conv. | |
2 × 2 Average pooling | |
Dense Block (2) | \(\left[ {\begin{array}{*{20}l} {1 \times 1\;{\text{ conv}}.} \\ {3 \times 3 \, \;{\text{conv}}.} \\ \end{array} } \right] \times 12\) |
1 × 1 conv. | |
2 × 2 Average pooling | |
Dense Block (3) | \(\left[ {\begin{array}{*{20}l} {1 \times 1 \, \;{\text{conv}}{.}} \\ {3 \times 3\;{\text{ conv}}.} \\ \end{array} } \right] \times 24\) |
1 × 1 conv. | |
2 × 2 Average pooling | |
Dense Block (4) | \(\left[ {\begin{array}{*{20}l} {1 \times 1\;{\text{ conv}}{.}} \\ {3 \times 3\;{\text{ conv}}{.}} \\ \end{array} } \right] \times 16\) |
7 × 7 Global average pooling | |
FC-2 | |
Soft-max | |
